w2v-bert-2.0-chichewa_34_34h

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the CLEAR-GLOBAL/CHICHEWA_34_34H - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.3084
• Wer: 0.3910
• Cer: 0.1127

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 100000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.2884	1.8458	1000	1.3872	1.0113	0.3808
0.092	3.6907	2000	0.5229	0.5397	0.1527
0.0604	5.5355	3000	0.4211	0.4785	0.1347
0.2837	7.3804	4000	0.3645	0.4376	0.1248
0.0217	9.2253	5000	0.3404	0.4469	0.1232
0.0299	11.0702	6000	0.3288	0.4160	0.1173
0.0162	12.9160	7000	0.3320	0.3983	0.1139
0.0436	14.7608	8000	0.3125	0.3847	0.1099
0.0205	16.6057	9000	0.3084	0.3910	0.1126
0.0198	18.4506	10000	0.4008	0.4002	0.1135
0.0516	20.2955	11000	0.3086	0.3701	0.1075
0.0057	22.1404	12000	0.3458	0.3847	0.1114
0.0041	23.9861	13000	0.3829	0.3899	0.1137
0.0142	25.8310	14000	0.4180	0.4121	0.1168

Framework versions

• Transformers 4.48.1
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1